paper-rag

Dashboard ​

/paper-rag:dashboard rebuilds a library-browser HTML app from the active home's current state — no hand-edited artifact to keep in sync.

/paper-rag:dashboard [--home PATH] [--home-name NAME] [--out DIR] [--open]

It scans every papers/*/metadata.json, checks which source files are present (PDF/XML/HTML, extracted figures), and sums papers/, chroma/, and bm25/ disk usage for the storage sidebar. It writes <home>/dashboard/index.html — a self-contained 3-pane app (icon rail / availability + keyword filters / sortable searchable reference table / detail rail with abstract, identifiers, and BibTeX export) plus dashboard/figs/<doc_key>.png cover images copied from each paper's first extracted figure.

The page also has: smart views (Unread / Starred / Needs attention), combinable availability + multi-select tag + year-histogram filters shown as removable chips, a first-author-highlighted Family G author column, per-paper reading status and stars, a hover info box on the amber ⚠ (open access, no PDF stored), Figures and Insights views, prev/next through the list (j/k, x select, s star, / search, Esc close), cite in BibTeX / APA / Vancouver / RIS, and export of the selection or the whole filtered view as .bib / .ris / .csv. Reading status, stars, and notes are stored in the browser (localStorage), not in the library.

Claims. Papers with extracted claims (see Asking Questions) show up in three places:

  • a Claims view (rail): every claim in the current filter with its direction, outcome, effect size and confidence interval, study design and paper, filterable by direction, study design and "effect size only". Selecting a claim shows its population / intervention / comparator / outcome, the effect, the verbatim evidence quote and its source section and page, with Open paper and Copy with PMID;
  • a Claims tab in each paper's detail pane (its own icon in the pane's toolbar, with a dot when the paper has claims), and a "N claims extracted" row in Data in library;
  • Insights cards for claim coverage and direction of findings, a "without claims" attention row, and a No claims smart view in the sidebar.

Each claim shows a source_level pill (abstract, fulltext or mixed) saying whether it was extracted from the abstract, the full text, or both. Claims stored before that field existed have no pill until /paper-rag:extract-claims is re-run for the paper.

The detail pane's toolbar groups its tabs (Details, Cite, PDF, Figures, Claims, Notes) in a segmented control, apart from the position counter and prev/next/close buttons.

Claim rows are read from Chroma metadata when the dashboard is built, so re-run /paper-rag:dashboard after /paper-rag:extract-claims. The chunk counts shown per paper exclude claim rows.

  • --home / --home-name — same home resolution as every other command (see Tags & Homes).
  • --out DIR — write elsewhere instead of <home>/dashboard.
  • --open — open the generated index.html in the default browser.

The script prints a JSON summary: out, n_papers, n_fulltext, n_figures, storage. Claude then publishes (or republishes, if config.json already has a dashboard_artifact_url) that file as an artifact, so re-running the command after an ingest updates the same link instead of creating a new one.

Run /paper-rag:dashboard again any time after /paper-rag:ingest to pick up new papers — there's no auto-refresh hook, since ingest is already slow.